Null vs. Alternative Hypothesis: What’s the Difference?

What Exactly Are the Null and Alternative Hypotheses?

You formulate a research hypothesis, collect your data, and reach the statistical analysis. Suddenly, two new statements appear: the null hypothesis, usually written as H₀, and the alternative hypothesis, commonly written as H₁ or H_a.

If your substantive expectation is that an intervention improves performance, why does the statistical procedure begin by considering a hypothesis of no improvement? And if you reject the null hypothesis, have you proved the alternative hypothesis?

The confusion often comes from treating null and alternative hypotheses as ordinary competing guesses. In classical statistical hypothesis testing, they have specific mathematical and inferential roles. Understanding those roles also helps prevent one of the most persistent errors in research reporting: interpreting failure to reject the null hypothesis as proof that nothing is happening.

The Null and Alternative Represent Competing Statistical Possibilities

How H0 and H1 Work in Statistical Hypothesis Testing

What Is the Null Hypothesis?

The null hypothesis specifies a particular claim about a population parameter or data-generating process against which the evidence is evaluated. In many familiar research applications, it represents no difference, no association, or no effect.

Suppose you compare mean examination scores between students receiving a new instructional intervention and students receiving conventional instruction. A simple null hypothesis might be expressed as:

H₀: μ_intervention = μ_comparison

Equivalently:

H₀: μ_intervention − μ_comparison = 0

Here, the null specifies that the population mean difference is zero.

However, "no effect" is not the universal definition of a null hypothesis. Depending on the statistical question, H₀ can specify another parameter value, a boundary, or a composite set of values. What matters is that it defines the condition against which the statistical test is constructed.

What Is the Alternative Hypothesis?

The alternative hypothesis specifies the competing parameter values or condition of interest. In a two-sided comparison of two means, it might be:

H₁: μ_intervention ≠ μ_comparison

This says that the population means differ, without specifying which one is larger.

If there is a justified directional prediction, the alternative could instead take a one-sided form:

H₁: μ_intervention > μ_comparison

That distinction matters because the alternative hypothesis helps determine whether the statistical test is one-sided or two-sided. The direction should be established from the research rationale rather than chosen after observing which way the sample results happen to point.

Null and Alternative Hypotheses at a Glance

Why Does Statistical Testing Begin With H0?

Classical significance testing asks how compatible the observed data, or something more extreme according to the test statistic, would be with a specified null model. A test statistic summarizes the discrepancy between the observations and what would be expected under H₀.

The p-value is then calculated under the assumption that the null hypothesis and the statistical model used for the test hold. Informally, it asks how unusual the observed test statistic, or one at least as extreme, would be under those assumptions.

If the p-value meets the prespecified significance criterion, researchers may reject H₀. If it does not, they fail to reject H₀.

This framework explains why statistical testing requires clearly specified competing hypotheses. NIST, for example, describes a statistical test as a mechanism for making quantitative decisions about a conjecture and notes that a statistical test requires null and alternative hypotheses.

Rejecting H0 Does Not Mean H0 Had Zero Probability of Being True

A p-value is not the probability that H₀ is true. Conventional frequentist significance testing does not calculate P(H₀ | data), the probability of the hypothesis given the observed data.

Instead, the calculation proceeds in the other direction: assuming H₀ and the relevant statistical assumptions, how unusual would the observed result or a more extreme one be?

This distinction is easy to reverse in ordinary language, but the two probabilities are not equivalent.

What Does Rejecting the Null Hypothesis Mean?

Suppose you set α =.05 before conducting the analysis and obtain a p-value of.012. Under the conventional decision rule, you would reject H₀ at the.05 significance level.

This means the observed evidence meets the criterion established for rejecting that null hypothesis under the statistical model. It does not establish that the alternative hypothesis is certainly true, that the effect is large, that the study is free from bias, or that the result will replicate.

Statistical significance addresses a narrower question than scientific importance.

What Does Failing to Reject H0 Mean?

Suppose instead that p =.18. Under a conventional α =.05 criterion, you fail to reject H₀.

The careful wording is fail to reject, not accept.

A nonsignificant result can occur because the null hypothesis is compatible with the data, but it can also occur when the study provides insufficient precision or statistical power to distinguish a meaningful effect from sampling variability. Measurement problems, small samples, heterogeneous effects, and model assumptions can also influence the result.

The Research Hypothesis and Statistical Alternative Are Related but Not Identical

Researchers sometimes use "alternative hypothesis" and "research hypothesis" interchangeably. They often correspond, but the concepts operate at different levels.

A research hypothesis is a substantive prediction about the phenomenon under investigation:

Students receiving retrieval practice will retain more material than students who reread the material.

A statistical alternative translates the relevant part of that substantive prediction into a statement about population parameters:

H₁: μ_retrieval > μ_rereading

Keeping those levels distinct helps explain why statistical and research hypotheses are not simply the same thing written differently.

Null and Alternative Hypotheses Must Fit Together

The hypotheses should define the relevant possibilities coherently. For a two-sided comparison of a population mean with a reference value μ₀, a familiar pair is:

H₀: μ = μ₀

H₁: μ ≠ μ₀

For a one-sided test, the hypotheses are formulated so that the null includes values outside the direction represented by the alternative. For example:

H₀: μ ≤ μ₀

H₁: μ > μ₀

The exact formulation depends on the parameter and test being used. This is why statistical hypotheses should not be written as decorative sentences disconnected from the planned analysis.

Directional Alternatives Change the Statistical Question

An alternative hypothesis stating μ₁ ≠ μ₂ asks whether the population means differ in either direction. An alternative stating μ₁ > μ₂ asks a narrower directional question.

That choice affects the rejection region in conventional hypothesis testing. It should therefore be determined before looking at the results and justified by the substantive research question. The underlying decision is closely connected to choosing between directional and nondirectional hypotheses.

Statistical Significance Is Not the Whole Scientific Conclusion

Even a correctly specified hypothesis test does not tell you everything you need to know. Researchers should also examine the estimated effect, its uncertainty, the quality of measurement, study design, assumptions, possible biases, substantive plausibility, and practical importance.

A tiny effect can be statistically detectable in a sufficiently large dataset. Conversely, a potentially important effect can remain statistically uncertain in a small or noisy study.

H₀ and H₁ therefore structure a statistical decision. They do not replace scientific interpretation.

How H0 and H1 Work in a Simple Group Comparison

Common Mistakes When Interpreting Null and Alternative Hypotheses

Write Statistical Hypotheses to Match the Question You Actually Intend to Test

Start with the substantive research question and determine what population parameter or model quantity represents it. Only then should you formulate H₀ and H₁ and choose an appropriate statistical procedure.

If you are still developing the substantive prediction rather than translating it into statistical notation, begin with the logic of developing a defensible research hypothesis. The statistical pair should follow from the research question and analysis, not substitute for them.

Before Testing Null and Alternative Hypotheses

Frequently Asked Questions About Null and Alternative Hypotheses

Rejecting H0 Is Not the Same as Proving H1

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.